通过启用提示词缓存将 Anthropic API 的代理心跳成本降低 90%。

💰 Running OpenClaw on a budget? Your biggest cost killer isn't the model—it's sending your full context with every single heartbeat.

48 heartbeats/day × 200K tokens = $240/month just keeping your agent "awake"

Quick win: Enable prompt caching in your Anthropic config. Cuts this by 90%.